Open Information Extraction via Contextual Sentence Decomposition

We show how contextual sentence decomposition (CSD), a technique originally developed for high-precision semantic search, can be used for open information extraction (OIE). Intuitively, CSD decomposes a sentence into the parts that semantically "belong together". By identifying the (implicit or explicit) verb in each such part, we obtain facts like in OIE. We compare our system, called CSD-IE, to three state-of-the-art OIE systems: ReVerb, OLLIE, and ClausIE. We consider the following aspects: accuracy (does the extracted triple express a meaningful fact, which is also expressed in the original sentence), minimality (can the extracted triple be further decomposed into smaller meaningful triples), coverage (percentage of text contained in at least one extracted triple), and number of facts extracted. We show how CSD-IE clearly outperforms ReVerb and OLLIE in terms of coverage and recall, but at comparable accuracy and minimality, and how CSD-IE achieves precision and recall comparable to ClausIE, but at significantly better minimality.